forensic identification computed

7/27/2019 forensic identification computed

http://slidepdf.com/reader/full/forensic-identification-computed 1/31

Forensic identification by computer-aided

craniofacial superimposition: a survey

SERGIO DAMAS, OSCAR CORDON and OSCAR IBANEZ

European Centre for Soft Computing

and

JOSE SANTAMARIA

University of Jaen

and

INMACULADA ALEMAN, MIGUEL BOTELLA, and FERNANDO NAVARRO

University of Granada

Craniofacial superimposition is a forensic process where a photograph of a missing person is com-pared with a skull found to determine its identity. After one century of development, craniofacialsuperimposition has become an interdisciplinary research field where computer sciences have ac-quired a key role as a complement of forensic sciences. Moreover, the availability of new digitalequipment (as computers and 3D scanners) has resulted in a significant advance in the applicabi-lity of this forensic identification technique. However, technology has also led to some confusion inforensic environments concerning the definition of a “computer-aided” system and an “automatic”method. The purpose of this contribution is twofold. On the one hand, we aim to clearly delimitthe different stages involved in the craniofacial superimposition process. Besides, we aim to clarifythe role played by computers in the methods considered in order to avoid the said confusion.

In order to accomplish these objectives, an up-to-date review of the recent works is presentedalong with a discussion of advantages and drawbacks of the existing approaches, with an emphasison the automatic ones. Upcoming case studies will be easily categorized by identifying which stageis tackled and which kind of computer-aided approach is chosen to face the identification problem.Remaining challenges are indicated and some directions for future research are given.

Categories and Subject Descriptors: J.3 [Life and Medical Sciences]: Medical information sys-tems; I.5.4 [Pattern Recognition]: Applications—Computer vision ; I.4.9 [Image Processing

and Computer Vision]: Applications; H.4.2 [Information Systems Applications]: Types of Systems—Decision support ; I.2.1 [Artificial Intelligence]: Applications and Expert Systems—Medicine and science

General Terms: Algorithms, Human Factors, Legal Aspects

Authors addresses: S. Damas, O. Cordon, and O. Ibanez, European Centre for Soft Computing, C.Gonzalo Gutierrez Quiros s.n., Edf. Cientıfico-Tecnologico, 33600 Mieres (Spain); J. Santamarıa,University of Jaen, C. Alfonso X El Sabio 28, 23700 Linares (Spain); I. Aleman, M. Botella, and F.Navarro, Physical Anthropology Lab, University of Granada, Av. de Madrid 11, 18012 Granada

(Spain). Corresponding email: [email protected] work is supported by the Spanish Ministerio de Educacion (ref. TIN2006-00829), and by theAndalusian Dpt. of Innovacion, Ciencia y Empresa (ref. TIC-1619), including EDRF fundings.Permission to make digital/hard copy of all or part of this material without fee for personalor classroom use provided that the copies are not made or distributed for profit or commercialadvantage, the ACM copyright/server notice, the title of the publication, and its date appear, andnotice is given that copying is by permission of the ACM, Inc. To copy otherwise, to republish,to post on servers, or to redistribute to lists requires prior specific permission and/or a fee.c 20YY ACM 0000-0000/20YY/0000-0001 $5.00

ACM Journal Name, Vol. V, No. N, Month 20YY, Pages 1–31.



2 · Sergio Damas et al.

Additional Key Words and Phrases: Forensic identification, craniofacial superimposition, photo-graphic supra-projection, skull-face superimposition, photographic superimposition, video super-imposition, skull modeling, decision making

1. INTRODUCTION

One of the main goals of forensic anthropology [Burns 2007] is to determine theidentity of a person from the study of some skeletal remains. In the last few decades,anthropologists have focused their attention on improving those techniques thatallow a more accurate identification.

Before making a decision on the identification, it is necessary to follow differentprocesses that let them assign a sex, age, human group, and height to the subjectfrom the study of bones found. Different methodologies have been proposed, ac-cording to the features of the different human groups of each region [Iscan 2005;Aleman et al. 1997; Gonzalez-Colmenares 2007].

Once the sample of candidates for identification is constrained by these preli-minary studies, an identification technique is applied. Among them, craniofacial superimposition [Krogman and Iscan 1986; Iscan 1993] is a complex and uncer-tain forensic process where a photograph of a missing person is compared with theskull that is found. Such comparison is guided by the proper correspondence of anumber of landmarks identified in both the skull (craniometric landmarks) and thephotograph (cephalometric landmarks).

Before reviewing the basis of this forensic identification technique, one shouldnote that different terms have been considered to refer to it during its more thanone century of development. This fact has been mainly due to the use of closesynonyms and specially to the coining of new, more specific terms depending on the

supporting technical devices considered through time. The next items justify ourchoice of “craniofacial superimposition” as the most general and currently extendedname for this forensic identification method.

—Craniofacial superimposition is the term widely found in the literature to referto all the tasks related to this forensic identification technique [Ubelaker et al.1992; Yoshino et al. 1995; Cattaneo 2007]. In particular, the most recent studiesconfirm the suitability of this terminology [Stephan 2009a; 2009b; 2009c; Ranson2009; Pickering and Bachman 2009].

—The term arises as a mean to differentiate between the forensic technique itself and the technical devices used to tackle the identification problem. Indeed, cra-

niofacial superimpositons were initially conducted using tracings made from pho-tographs [Webster 1955; Sen 1962] and authors refereed to the procedure as “pho-tographic superimposition” [Dorion 1983; Brocklebank and Holmgren 1989; Maat1989]. Because of the rapid developments in video technology, the term “videosuperimposition” was later used when this tool became common in forensic iden-tification [Seta and Yoshino 1993; Pesce Delfino et al. 1993; Shahrom et al. 1996;Yoshino et al. 1997]. Finally, the use of computers to assist the anthropologistsin the identification process involved the next generation of craniofacial superim-

ACM Journal Name, Vol. V, No. N, Month 20YY.



Forensic identification by computer-aided craniofacial superimposition: a survey · 3

position systems1. The latter approaches are usually referred to as “computer-aided” or “computer-assisted craniofacial superimposition”2 [Pesce Delfino et al.

1986; Ubelaker et al. 1992; Aulsebrook et al. 1995; Yoshino et al. 1997]. Thecurrent review will be devoted to these kinds of systems and specifically to themost advanced ones based on the use of computer vision, 3D modeling and ma-chine learning-based automatic methods. These systems have not been carried aspecific distinctive name till now despite the fact of being fundamental tools inthe computer-aided procedure nowadays.

—Hence, when using the generic term “craniofacial superimposition” we are assu-ming neither a particular acquisition device nor a given data format as the inputsof our problem. We just consider that any craniofacial superimposition methodwill deal with a 2D image of the disappeared person (typically a photograph)and the skull found (maybe as a part of other skeletal remains).

—There are some sources that use the term “photographic supra-

projection” [Bronkhorst 2006; Stratmann 1998] instead. We avoid its usebecause it does not explicitly indicate that a matching of a skull with a face isspecifically involved.

—Finally, as we will show in Section 2, craniofacial superimposition should not bemislead by the second stage of this forensic technique, i.e. the skull-face overlay,since it refers to the whole identification process including the data acquisitionand processing as well as the decision making stages. On the other hand, cranio-facial superimposition should not be mislead by craniofacial identification either.Notice that, the latter term is used as an umbrella including both craniofacialsuperimposition and facial approximation3 [Clement and Ranson 1998; Stephan2009b; Wilkinson 2009a]. Both methods are underpinned by knowledge of humancraniofacial anatomy. It is this principle which ties these two techniques together

despite the use of different technical protocols for each of them.

Successful comparison of human skeletal remains with artistic or photographicreplicas has been achieved many times using the craniofacial superimposition tech-nique, ranging from the studies of the skeletal remains of the poet Dante Alighieriin the nineteenth century [Welcker 1867], to the identification of victims of therecent Indian Ocean tsunami [Al-Amad et al. 2006]. Among the huge number of case studies where craniofacial superimposition has been applied4, it was helpful in

1Attempts to achieve high identification accuracy through the utilization of advanced computertechnology has been a monumental task for experts in the field in the last two decades [Lan 1992].2Notice that, the terms “skull-face superimposition”, “skull-photo superimposition”, “photograp-hic superimposition” or “video superimposition” have also been used in combination with the

“computer-aided/assisted” adjective.3“In the past, facial approximation methods have been known by many other names. The mostpopular of these is facial reconstruction . This name is strongly misleading as it leaves the erroneousimpression that the methods are exact, reliable and scientific” [Stephan 2009a]. Besides, Lan et.al reported in 1992 that their system had been used to identify more than 300 cases in China bythat time [Lan 1992]4Although there is not a register to evaluate the exact number of cases in which craniofacialsuperimposition has been used and/or resulted in positive results, it would appear to be in thehundreds only in Australia [Stephan et al. 2008].





the identification of well-known criminals as John Demanjuk (known to Nazi con-centration camp survivors as “Ivan the Terrible”) and Adolf Hitler’s chief medical

officer Dr. Josef Mengele at Sao Paulo, Brazil in 1985 [Helmer 1988]. It is also usedin the identification of terrorists nowadays [Indriati 2009].

Important contributions during the first epoch of craniofacial superimpositionare those devoted to study the correspondence of the cranial structures with thesoft tissue covering them [Broca 1875]. Bertillon [1896] introduced the basis tocollect physiognomic data of the accused of a crime at the end of the nineteenthcentury. Such data is still used nowadays. Much later, Martin and Saller [1966]proposed the basis to systematize this discipline. Following those premises, theusual procedure of the first identifications by means of craniofacial superimpositionconsisted of obtaining the negative of the original face photograph and markingthe cephalometric landmarks on it. The same task was to be done with a skullphotograph. Then, both negatives were overlapped and the positive was developed.

The technological support for the technique from these initial identifications in-volved a large number of very diverse approaches found in the literature. Thatcould also be the reason for the current diversity of craniofacial superimpositionmethods and their terminology, as mentioned before. Instead of following a uni-form methodology, every expert tends to apply his own approach to the problembased on the available technology and on his deep knowledge on human craniofacialanatomy, soft tissues, and their relationships. Therefore, craniofacial superimposi-tion approaches evolved as new technology was available although their foundationswere previously laid.

Some of these approaches were classified in a review by Aulsebrook et. al [1995]according to the technology used to acquire the data and to support the skull-faceoverlay and identification processes, i.e. static photographic transparency, video

technology, and computer graphics. Similar classification schemes have been alsoreported by other authors [Nickerson et al. 1991; Yoshino and Seta 2000], whichdescribe how craniofacial superimposition has passed through three phases: photo-graphic superimposition (developed in the mid 1930s), video superimposition (wi-dely used since the second half of the 1970s), and computer-aided superimposition(introduced in the second half of the 1980s). Moreover, Yoshino et al. [1997] clas-sified some of the computer-aided craniofacial superimposition methods into twocategories from the viewpoint of the identification strategy. The first strategy is todigitalize the skull and facial photographs and then morphologically compare thetwo images by image processing. The second is to evaluate the fit between the skulland facial image by morphometric examination.

Notice that, the latter contributions are previous to the image processing boom of the last decade. Indeed, important issues like 3D modeling and machine learning areneglected. In case it is used, the computer is usually considered just as a secondarysupport for the technique even when authors claim they followed a “computer-aided” approach [Ubelaker et al. 1992; Ricci et al. 2006].

The aim of this survey is to update previous reviews, both including differentworks published afterwards and considering a new computing-based classificationcriterion. That criterion is more related to the use of computers in the differentstages of the craniofacial superimposition process itself. The different stages in-





volved in the craniofacial process will be thus clearly delimited. In our opinion,to properly characterize any craniofacial superimposition system (and specifically

computer-aided ones), the whole process should be considered as divided up intothree consecutive stages, namely face enhancement and skull modeling, skull-faceoverlay, and decision making. We will point out advantages and disadvantages of different approaches, with an emphasis on the computer-aided techniques that have been employed and on the tasks these techniques are supposed to solve in an auto-matic manner. We are interested in the methods, not on the analysis of specificcases. Hence, papers reporting only case studies will be out of the scope of thissurvey.

We should especially remark that we will not judge the effectiveness of the met-hods due to the unavailability of the tackled cases and used equipments, and there-fore the impossibility to reproduce the experimental setup and to perform compa-rative experiments. As stated by Carl N. Stephan, “presently, it is not possible to

draw firm statements concerning the overarching performance of superimpositionmethods because formal published studies on the accuracy and reliability of themethods have been infrequent, have used small samples, and have often not beenreplicated” [Stephan 2009a].

The current proposal is organized as follows. In Section 2 we will give an over-view of the craniofacial superimposition process with a brief description of thestages that compose it. In Section 3 we will present the role performed by the com-puter to accomplish every craniofacial superimposition stage. We will review andcategorize the existing contributions in Section 4. Some works partially related tothe craniofacial superimposition process will be shortly listed in Section 5. Finally,Section 6 will be devoted to a discussion of solved and not solved problems, trends,and challenges for future research.

2. THE CRANIOFACIAL SUPERIMPOSITION PROCESS

As said, craniofacial superimposition [Iscan 1993] is a forensic process where pho-tographs or video shots of a missing person are compared with the skull that isfound. By projecting both photographs on top of each other (or, even better,matching a scanned three-dimensional skull model against the face photo/series of video shots), the forensic anthropologist can try to establish whether that is thesame person [Krogman and Iscan 1986].

The said process is guided by a number of landmarks located in both the skulland the photograph of the missing person. The selected landmarks are located inthose parts where the thickness of the soft tissue is low. The goal is to ease theirlocation when the anthropologist must deal with changes in age, weight, and facialexpressions.

The skull landmarks [George 1993] that are often used (see Figure 1) follow:

. Craniometric landmarks:

. Dacryon (Da): The point of junction of the frontal, maxillary, and lacrimalbones on the lateral wall of the orbit.

. Frontomalar temporal (Fmt): The most lateral point of junction of the frontaland zigomatic bones.





Fig. 1. From left to right, principal craniometric landmarks: lateral and frontal views

. Glabella (G): The most prominent point between the supraorbital ridges in themidsagittal plane.

. Gnathion (Gn): A constructed point midway between the most anterior (Pog)and most inferior (Me) points on the chin.

. Gonion (Go): A constructed point, the intersection of the lines tangent to theposterior margin of the ascending ramus and the mandibular base, or the mostlateral point at the mandibular angle.

. Nasion (N): The midpoint of the suture between the frontal and the two nasalbones.

. Nasospinale (Ns): The point where a line drawn between the lower margins of the right and left nasal apertures is intersected by the midsagittal plane (MSP).

. Pogonion (Pog): The most anterior point in the midline on the mental protu-berance.

. Prosthion (Pr): The apex of the alveolus in the midline between the maxillarycentral incisor.

. Zygion (Zy): The most lateral point on the zygomatic arch.

Meanwhile, the most usual face landmarks (see Figure 2) are:

. Cephalometric landmarks:

. Alare (al): The most lateral point on the alar contour.

. Ectocanthion (Ec): The point at the outer commissure (lateral canthus) of thepalpebral fissure just medial to the malar tubercle (of Whitnall) to which the lateralpalpebral ligaments are attached.

. Endocanthion (En): The point at the inner commissure (medial canthus) of thepalpebral fissure.

. Glabella (g’): In the midline, the most prominent point between the eyebrows.

. Gnathion (gn’): The point on the soft tissue chin midway between Pog and Me.

. Gonion (go’): The most lateral point of the jawline at the mandibular angle.

. Menton (Me): The lowest point on the MSP of the soft tissue chin.

. Nasion (n): In the midline, the point of maximum concavity between the noseand forehead. Frontally, this point is located at the midpoint of a tangent between





Fig. 2. From left to right, principal facial landmarks: lateral and frontal views

the right and left superior palpebral folds.

. Pogonion (pog’): The most anterior point of the soft tissue chin.

. Labiale inferius (Li): The midpoint on the vermilion line of the lower lip.

. Labiale superius (Ls): The midpoint on the vermilion line of the upper lip.

. Subnasale (sn): The midpoint of the columella base at the angle where thelower border of the nasal septum meets the upper lip.

. Tragion (t) Point in the notch just above the tragus of the ear; it lies 1 to 2mm below the spine of the helix, which can be palpated.

. Zygion (Zy’): The most lateral point of the check (zygomaticomalar) region.

Therefore, in every system for skull identification by craniofacial superimpositiontwo objects are involved: a skull and a face image. The latter is typically a photo-graph although it can be sometimes replaced by a series of video shots or, in few

cases, a portrait of the missing person. The final goal, common to every system,is to assess the anatomical consistency between the skull and the face. Differenttechnologies and methods are used to achieve such goal, starting from the acquisi-tion of the input data, that could be either to take a photograph of the skull in thesame orientation of the face, to acquire a video of both skull and face with the helpof a mixing device, or to acquire a 3D model of the skull and a digital photograph.Then, the methods to superimpose the acquired data range from the use of slidesto the application of computer graphics techniques. Likewise, several methods of identification can be employed, sometimes related to the technology used to acquirethe data, and other times independently chosen.

Readers interested in details about photographic and video superimposition sys-tems can refer to [Yoshino and Seta 2000]. In this survey we will focus on computer-aided techniques, that we believe to be faster and more objective compared with

manual and visual ones. However, there is some confusion in the forensic litera-ture concerning the definition of a “computer-based” system and an “automatic”method. We will try to clarify that issue in Section 3.

In our view, the whole craniofacial superimposition process is composed of threestages (see Figure 3):

(1) The first stage involves achieving a digital model of the skull and the en-hancement of the face image. This stage is not present in all the systems. Indeed,





Craniofacial superimposition

skull−face overlay

decision making

face enhancement &

skull modeling

Fig. 3. The three stages involved the craniofacial superimposition process

the oldest systems and most of the recent ones still acquire a photograph and/orvideo shots of the skull, instead of building a 3D model of it5. As we will explain inSection 6.1, obtaining an accurate 3D cranial model has been considered a difficulttask by forensic anthropologists in the past. However, it is nowadays an affordableand attainable activity using laser range scanners (Figure 4) like the one used byour team, available in the Physical Anthropology Lab at the University of Granada(Spain) (Santamarıa et al. [007a; 007b; 009a]). The subject of the identificationprocess, i.e. the skull, is a 3D object. Hence, the use of a skull 3D model instead of a skull 2D image should be preferred because it is definitively a more accurate re-

presentation. It has already been shown that 3D models are much more informativein other forensic identification tasks [De Angelis et al. 2009]. Concerning the faceimage, the most recent systems use a 2D digital image. This stage aims to applyimage processing techniques [Gonzalez and Woods 2008] in order to enhance thequality of the face image that was typically provided when the person disappeared.

(2) The second stage is the skull-face overlay. It consists of searching for thebest overlay of either the skull and face 2D images or the skull 3D model and theface 2D image achieved during the first stage. This is usually done by bringing tomatch some corresponding landmarks on the skull and the face.

(3) Finally, the third stage of the craniofacial superimposition process corres-ponds to the decision making. Based on the skull-face overlay achieved, the identi-fication decision is made by either judging the matching between the corresponding

landmarks in the skull and in the face, or by analyzing the respective profiles. No-tice that, the use of computers in this stage aims to support the final identificationdecision that will be always made by the forensic anthropologist.

As can be seen, the whole procedure is very time consuming. As said, there isnot a systematic methodology but every expert usually applies a particular process.

5In that case, the first stage should be better called “face and skull data acquisition.”





Fig. 4. Acquisition of a skull 3D partial view using a Konica-MinoltaTMlaser range scanner

Hence, there is a strong interest in designing automatic methods to support theforensic anthropologist to put it into effect [Ubelaker 2000]. Moreover, the accuracyof each step of the process will influence the subsequent one. Thus, there is a needfor optimizing each individual step properly.

3. COMPUTER-AIDED CRANIOFACIAL SUPERIMPOSITION

As said, the differentiation between methods that do not use computer techno-logy and those that use it has already been proposed [Aulsebrook et al. 1995]. Inthe literature, photographic and video superimposition have been considered tobelong to the former category. Meanwhile, methods defined as digital or computer-aided craniofacial superimposition techniques have been considered to belong tothe latter. Thus, the distinction between computer-aided and non-computer-aidedmethods has been clearly guided by the use of computer-based technology alongthe craniofacial superimposition process up to now. Nevertheless, the role of thecomputer in that process is really important nowadays and it was not consideredin previous reviews. Moreover, the analysis of previous contributions is especiallydifficult when some authors claim they propose a “computer-aided” or “‘computer-assisted” system [Ricci et al. 2006] and the computer mainly plays the role of asimple visualization tool.

Hence, to fill that gap, we will expand the computer-aided category defined inprevious reviews by distinguishing between non-automatic and automatic methods.Computer-aided non-automatic methods use some kind of digital infrastructure tosupport the craniofacial superimposition process, i.e. computers are used for storingand/or visualizing the data. However, they are characterized by the fact that their

computational capacity to automate human tasks is not considered. On the otherhand, computer-aided automatic methods use computer programs to accomplishan identification sub-task itself. There are some remarks that should be doneconcerning the three process stages:

(1) Regarding the first stage, automatic methods may deal with either the 2Dface image or the skull. On the one hand, when dealing with the 2D face image,automatic systems accomplish the restoration of the photograph by means of digital





image processing techniques. On the other hand, the aim of automatic methodsconcerning the skull is the achievement of an accurate 3D model.

(2) Concerning the second stage, we will point out a clear division betweencomputer-aided non-automatic and automatic skull-face overlay methods. The for-mer ones use computers to support the overlay procedure and/or to visualize theskull, the face, and the obtained superimposition. Nevertheless, the size and orien-tation of the skull is changed manually to correctly match that one of the headin the photograph. This is achieved by either physically moving the skull, whilecomputers are simply used to visualize it on the monitor, or (with the help of somecommercial software) by moving its digital image on the screen until a good matchis found. On the opposite, the latter ones, i.e. automatic skull-face overlay met-hods, find the optimal superimposition between the 3D model of the skull and the2D face image using computer programs.

(3) Finally, regarding the decision making stage, automatic systems assist theforensic expert by applying decision support systems [Keen 1978]. Moreover, thosecomputer programs must use objective and numerical data for evaluating the ob-tained matching between the skull and the face. Based on that evaluation, thesystem suggests an identification decision to the forensic expert. Thus, the decisionsupport system is intended to help decision makers compile useful information fromthe analysis of the skull-face overlay outcomes. Of course, the final decision will bealways made by the anthropologist according to both the support of the automaticsystem and his expertise. On the other hand, if the identification decision only re-lies on the human expert who visually evaluates the skull-face overlay obtained inthe previous stage, then the method will be considered as a non-automatic system,although it might use digital data as a supporting means.

4. CLASSIFICATION AND DISCUSSION OF EXISTING WORKS

In this section we will review and categorize the existing contributions of computer-aided craniofacial superimposition systems. They will be classified according to thestage of the process which is addressed using a computer-aided method. Informationabout the method used for the remaining stages will be given shortly together witha brief discussion.

Unfortunately, these stages are not so clearly distinguished in some of the existingcraniofacial superimposition methods as we might expect. This fact causes someconfusion as sometimes authors themselves define their own method as computer-

aided craniofacial superimposition when they refer only to the decision makingstage and others refer to identification method when they tackle the skull-faceoverlay stage. That is one of the reasons why our categorization is different fromthe previous ones that can be found in the literature.

Table I gives an overview of the papers describing computer-aided systems exa-mined in the present survey. Studies are listed in chronological order. Additionalinformation about the input data needed and/or the “classic” superimposition met-hod used are included, when available.





Table I. An overview of the literature on computer-aided forensic identification

systems by craniofacial superimposition. The stage of the process, i.e. skull mo-deling (SM), skull-face overlay (SF) and decision making (DM), that is addressedusing a computer-aided method is indicated with CA (computer-aided automaticmethods), or CN (computer-aided non-automatic methods). Notice that, if oneparticular stage is not tackled using computers it is noted by NC.

SM SF DM Remarks

[Lan and Cai 1985] CN uses “digital” superimposition[Tao 1986] CN CA uses “digital” superimpositionLan et al. [1988; 1990; 1993] CN CA uses “digital” sup erimp ositionPesceDelfino et al.[1986; 1993] CA manual p ositioning of the skull[Nickerson et al. 1991] CA CA binary-coded genetic algorithm[Ubelaker et al. 1992] CN CA uses “digital” superimposition[Bajnoczky and Kiralyfalvi 1995] CA based on video superimposition

Yoshino et al. [1995; 1997] CN CA photo-video superimposition[Shahrom et al. 1996] CN 3D model in facial approximationGhosh and Sinha [2001; 2005] CA works on 2D skull images[Scully and Nambiar 2002] CN works on 2D skull images[Bilge et al. 2003] CN NC based on commercial software[Biwasaka et al. 2005] CA based on optical techniques[Al-Amad et al. 2006] CN NC based on commercial software[Galantucci et al. 2006] CN computed tomography vs. laser[Ricci et al. 2006] CN CA based on 2D skull radiographsSantamarıa et al. [007b; 007a; 009a] CA adjancent overlapping regions[Ballerini et al. 2007] CA real-coded genetic algorithm[Fantini et al. 2008] CN based on commercial softwareIbanez et al. [2008; 2009b] CA uncertain landmark location

[Ibanez et al. 2009a] CA real-coded evolutionary algorithms[Benazzi et al. 2009] CN based on commercial software[Ballerini et al. 2009] CA use of heuristic features[Santamarıa et al. 009b] CA coplanar landmarks avoidance

4.1 Face enhancement and skull modeling

Let us highlight the difference between the face image and the skull model. Theface image is typically a photograph. It was acquired under some conditions thatare fixed and usually unknown at the moment of the forensic analysis. The onlypossibility is to attempt to enhance its quality. If it is not in digital format, it can bescanned and transformed into a 2D digital image. Then, it can be enhanced using

digital image filters and/or processing algorithms. On the other hand, the skull isan available physical object and its model needs to be obtained to accomplish anautomatic procedure.

We will detail both face enhancement and skull modeling procedures. On the onehand, a good quality of the face image is needed [Nickerson et al. 1991]. Thereforeenhancement techniques are applied [Gonzalez and Woods 2008]. Such techniquesdepend on the available format (digital camera image or scanned photographic pa-per) and include frequency domain filters to fix artifacts due to aliasing and sam-





pling problems present in scanned documents, as well as removal of non uniformillumination effects and sharpening methods to deal with blurring and problems

related to movements. Notice that, the proper filter and its most suitable parame-ters are a choice that must be performed by the expert since they highly dependon the aforementioned acquisition conditions. According to Section 3, approachesthat use human operated commercial software for the 2D face image enhancementwill be considered non-automatic methods. Automatic methods perform such 2Dimage enhancement using computer programs.

On the other hand, recent techniques for craniofacial superimposition need anaccurate 3D model of the skull. The possibilities of recording 3D forensic objects arenot so many considering the available resources of a typical forensic anthropologylab. In the biomedical field computed tomography scanning images are the startingdata to reconstruct the skull [Singare et al. 2009; Fantini et al. 2008]. However,many forensic labs are exploiting the capabilities of laser range scanners nowadays.

That is due to the fact that these devices present a greater availability and alower cost. Thus, we will focus our skull modeling study on the contributions thatuse laser range scanners instead of other devices that have also been consideredto obtain a 3D model of the skull in other application domains [Nakasima 2005;Enciso et al. 2003]. Laser range scanners are based on the optical principle of triangulation and acquire a dense set of three-dimensional point data in a very rapid,non-contact fashion [Bernardini and Rushmeier 2002]. In lucky cases, laser rangescanners are equipped with an additional positioning device named rotary tableand appropriate software that permits the 3D reconstruction. Nevertheless, thereare situations where that software do not provide suitable 3D models. Moreover,there are scenarios where it is not even possible to use a rotary table.

Before going on with the 3D modeling process, every 3D view of the skull acquired

by the laser range scanner must be preprocessed. This task involves the cleaning,smoothing, and filling of the view. Cleaning aims to remove those artifacts thatwere acquired by the scanner as part of the scene but which do not correspondto the skull. Meanwhile, smoothing is mainly concerned with the removal of someartificial vertices that could have been wrongly included by the scanner on theborders of the surface because of a perspective distortion. Fortunately, this task isnot needed so often. Finally, filling is used to avoid small holes to appear in thoseparts of the skull that are not properly scanned because they are too dark for thescanner capabilities or they are located in shadow regions.

In order to accomplish the 3D model some anthropologists are skilled enough todeal with the set of 3D views and they supervise the procedure of a commercial soft-ware like RapidFormTM. Sometimes, this software does not provide the expected

outcomes and the anthropologists even have to stitch up manually every couple of adjacent views. Hence, 3D image reconstruction software is a real need to constructthe 3D model by aligning the views in a common coordinate frame. Such processis usually referred as range image registration [Brown 1992; Ikeuchi and Sato 2001;Zitova and Flusser 2003]. It consists of finding the best 3D rigid transformation(composed of a rotation and a translation) to align the acquired views of the object.An example of three different views of a skull and the reconstructed 3D model isshown in Figure 5.





Fig. 5. Three different views of a skull and the reconstructed 3D model

In this section we will mainly focus on contributions that include an automatic 3Dmodeling procedure because all the rest of the methods do not consider this stageand directly acquire a 2D projection of the skull (i.e., a skull photo). According toSection 3, all the approaches that use computers but do not consider the 3D skullmodel will be considered non-automatic methods [Yoshino et al. 1995; Ghosh andSinha 2001; Pesce Delfino et al. 1986; Ricci et al. 2006].

Up to our knowledge, Nickerson et al. [1991] were the first researchers to proposethe use of a 3D model to tackle the craniofacial superimposition problem. In theirwork, a range scanner and a digital camera were used for 3D digitalization of the

skull surface mesh and the 2D antemortem facial photograph, respectively. Wellknown image processing algorithms were used for image enhancement (median fil-tering, histogram equalization, Wiener filtering) [Gonzalez and Woods 2008]. Ren-dering was done through computer graphics techniques. A feature-based algorithmto reduce the computational and memory complexities inherent in solid modelingis also described.

Shahrom et al. [1996] followed a similar approach based on the use of a 3D laserrange scanner. Authors used a skull holder, which could be slowly rotated through360◦ in a horizontal plane under computer control. The 3D model was later usedin facial approximation.

A completely different approach is presented in [Biwasaka et al. 2005] where theauthors examined the applicability of holography in the 3D recording of forensic

objects. Holography is known as a unique optical technique capable of recordingthe 3D data of an object. Two types of images, real and virtual, can be obtainedfrom a holographically exposed film, or hologram. Two superimposition systemsusing holographic images were examined in order to evaluate the potential use of this recording method. The authors claim that the performance of holography iscomparable to that of the computer graphics system, which consists of an imagescanner, software, and a display unit. Moreover, they argue it can even be superiorto the computer technique with respect to the 3D reconstruction of images. We





believe in the potential of this technique. However, the use of an automatic super-imposition method and a comparison with a reconstructed 3D range image could

have objectively proved the actual utility of holography in this field.Galantucci et al. [2006] compared two different acquisition techniques of images

of a skull. In particular, computed tomography and laser range scanners perfor-mance are compared to ascertain which enabled more accurate reproductions of the original specimen. Comparison between the original and every model yieldedsatisfactory results for both techniques. However, computed tomography scanningdemonstrated some advantages over the laser technique, as it provided a cleanerpoint cloud, enabling shorter pre-processing times, as well as data on the internalparts, which resulted in the reproduction of a more faithful model.

Santamarıa et al [007b; 007a; 009a] proposed a method, based on evolutionaryalgorithms [Back et al. 1997], for the automatic alignment of skull range images.Different views of the skull to be modeled were acquired by using a laser range

scanner. A two step pair-wise range image registration technique was successfullyapplied to such images. The method includes a pre-alignment stage, that uses ascatter-search based algorithm [Laguna and Martı 2003], and a refinement stagebased on the classical iterative closest point algorithm [Besl and McKay 1992]. Themethod is very robust, indeed it reconstructs the skull 3D model even if there is noturn table and the views are wrongly scanned.

Fantini et al. [2008] used a laser range scanner to create a 3D model of a medievaldamaged skull. The large missing part of the skull allowed scanning both outer andinner surfaces of the ob ject. Thirty three partial views were needed to completethe acquisition of the whole surface by rotating the skull. Through post-processingof the data collected from the 3D scans, a triangular mesh was finally obtained.Those operations were performed by RapidForm 2006, RE TMcommercial software.

A similar approach was followed in [Benazzi et al. 2009] in order to tackle the 3D

skull reconstruction of Dante Alighieri (1265-1321) as part of a project to achievethe facial approximation of the famous poet. Based on the data provided by a laserrange scanner, the model of Dante’s skull was constructed using the utilities provi-ded by the Rapidform XOS2TMcommercial software. In particular, authors refer tooperations as registration and merging of the point clouds, and simplification andediting of the digital model.

Ballerini et al. [2009] proposed the automatic reduction of the data provided bythe laser range scanner used in the skull 3D model reconstruction task. The densepoint cloud corresponding to every skull view is synthesized by considering heuristicfeatures that are based on the curvature values of the skull surface. Those featuresguide the automatic 3D skull model reconstruction by means of an evolutionaryalgorithm.

4.2 Skull-face overlay

The success of the superimposition technique requires positioning the skull in thesame pose of the face as seen in the given photograph. The orientation processis a very challenging and time-consuming part of the craniofacial superimpositiontechnique [Fenton et al. 2008]. Most of the existing craniofacial superimpositionmethods are guided by a number of landmarks of the skull and the face (see Sec-tion 2). Once these landmarks are available, the skull-face overlay procedure is





based on searching for the skull orientation leading to the best matching of the setof landmarks. Scientific methods for positioning the skull had already been propo-

sed before computers become largely available [Chandra Sekharan 1993; Glaisterand Brash 1937; Kumari and Chandra Sekharan 1992; Maat 1989].

These methods are not computer-aided but are somehow closer to them than totrial and error procedures. In these approaches the skull is manually placed on atripod. However, its pose is estimated using a mathematical procedure, instead of atrial and error routine. The researchers applying these methods calculate the headsize and orientation in the photograph, so they can position the skull in the sameposture. We briefly summarize those contributions as follows:

—In very early approaches [Glaister and Brash 1937] the enlargement factor iscalculated based on linear measurements of items within the antemortem photo-graph, such as fabric, button, tie, and other objects of known geometry (doors,chairs, etc.) [Chandra Sekharan 1993]. Other scale correlation methodology has

included measurement of the interpupillary distance and size of dentition [Austin-Smith and Maples 1994].

—Maat [1989] proposed to use a set of anthropometrical landmarks, along with rela-tive reference lines, to calculate the three components of head rotation (“bendingforward”, “turning sideways”, and “rolling sidewards”) to posture the skull. Theprinciple of central projection and a minimum photographic distance of 1.5 m

are important preconditions.

—Chandra Sekharan [1993] suggested using the vertical distance “d” between theectocanthions and tragion as a measure for calculating the extent of flexion orextension of the head. The extent of the rotation of the face was calculated fromthe L/R ratio, where L and R denote the distances between the left and rightectocanthion from the midline of the face. Using these factors, the skull underexamination was positioned on a tripod stand with the help of a remote controlpositioning device [Kumari and Chandra Sekharan 1992]. A practical suggestionfor the camera distance was also given.

However, these methods are out of the scope of this proposal that is focused oncomputer-aided skull-face overlay contributions. Within this group of approacheswe will differentiate between non-automatic and automatic works as follows.

4.2.1 Non-automatic skull-face overlay methods. Below, we describe skull-faceoverlay methods known as computer-aided methods in the literature. Nevertheless,we prefer to refer to them as computer-aided non-automatic skull-face overlay met-hods. They are typical examples of the use of a digital infrastructure but withouttaking advantage of its potential utility as automatic support tools for the foren-

sic anthropologists. Notice that, they depend on good visualization and overlaymechanisms to aid human operators. Hence, processes following this approach areprone to be time-consuming, hard to be reproduced, and subjective.

—[Lan and Cai 1985] developed a craniofacial superimposition apparatus calledTLGA-1, based on the principles of dual projection. During the following years,these authors evolved this system resulting in new subsequent versions, TLGA-2 and finally TLGA-213 [Tao 1986; Lan and Cai 1988; Lan 1990; Lan and Cai





1993]. The TLGA-213 system was mainly composed of a TV camera, a computer,an A/D and D/A converter, a mouse, and the 213 system software library. The

system calculated the pitch angle of the face photograph by measuring the ratiobetween the distances in the vertical line segments glabella to nasion and gnathion to nasion . The natural head size was calculated from the distance between theectocanthions and the deflection angle in the photograph. The latter parametersare iteratively computed and considered as a guide for the manually performedskull-face overlay.

—Ubelaker et al. [1992] solved a huge number of cases submitted to the SmithsonianInstitute by the FBI. Their software allows any desired combination of skeletal-photograph comparison, including the chance to remove the soft tissue to viewthe underlying skeletal structure. It works on digitalized images of both faceand skull and offers the possibility to assess the consistency between them. Theidentification procedure usually requires less than one hour. It is not specified if

this time includes the acquisition and skull-face overlay steps or only the decisionmaking stage. However for the acquisition of the digital images, the authorsvisualize the facial photograph and trace anatomical landmarks on a plastic slidetaped on the monitor. Then, they visualize the skull and manually manipulateit to match the marked landmarks. The quality of the photograph and theproper orientation of the skull are claimed to highly influence the success of thetechnique.

—Yoshino et al.’s skull identification system [Yoshino et al. 1997] consists of twomain units, namely a video skull-face overlay system and a computer-aided de-cision making system. In the former, the determination of the orientation andsize of the skull to those of the facial photograph is done by a pulse motor-drivenmechanism, through the help of the fade-out or wipe mode of the video image

mixing device. Then, the skull and facial images are digitalized, stored in thecomputer, and superimposed on the monitor.

—Ricci et al. [2006] presented an algorithm to compare a facial image with a skullradiograph. Thus they work with pairs of 2D images and the superimposition isdone by the human operator that manually marks anatomical points and bringsthem to match. Their software seems to account only for translation and scaling,while the algorithm is able to compensate for up to 10o of head rotation. However,the algorithm only calculates distances and threshold in an automatic way, whilethe skull-face overlay is done manually.

—The use of commercial software as Adobe PhotoshopTMhas been reported byBilge et al. [2003] and Al-Amad et al. [2006]. They use the “free transform”tool to adjust the scale of the face photograph, superimposed over the skull

photo. The “semi-transparent” utility allows the operator to see both imageswhile moving, rotating, and resizing the overlaid image (see Figure 6). A similarapproach was also used in both [Scully and Nambiar 2002] and [Ricci et al. 2006]to validate a classical method and to superimpose skull radiographs, respectively.

4.2.2 Automatic skull-face overlay methods. We have found only few really in-teresting works to perform skull-face overlay in a fully automatic way. They arebased on the use of machine learning algorithms [Mitchell 1997] from artificial





Fig. 6. Non-automatic skull-face overlay based on PhotoshopTM

intelligence, as artificial neural networks [Rumelhart and McClelland 1986], evolu-tionary algorithms [Back et al. 1997], and fuzzy logic [Zadeh 1965]. The automationprovided by these approaches represents an added value since they are typically fas-ter than non-automatic methods. Moreover, they rely on quantitative measures andthey can be easily reproduced. However, this sort of works often involve technicalconcepts that are usually unknown by most of the forensic anthropologists. Thus,a multidisciplinary research team is required. A brief description of the methods inthis group is provided as follows:

—The method proposed by Ghosh and Sinha [2001] is an adaptation of their pre-vious work for face recognition problems [Sinha 1998] and it was recently applied

to an unusual case [Ghosh and Sinha 2005]. The Extended Symmetry PerceivingAdaptive Neuronet (ESPAN) consists of two neural networks to be applied to twodifferent parts of the overlaying and allows to select fuzzy facial features to ac-count for ambiguities due to soft tissue thickness. More in details, the system canimplement an objective assessment of the symmetry between two nearly front 2Dimages: the cranial image and the facial image that are the inputs as the sourceand the target images, respectively. The output is the mapped cranial imagesuitable for superimposition. Two neural networks need to be trained separatelybecause each of them can correctly map only a part of the cranial image. Twolimitations are pointed out by the authors: i) a part of the cranial image will notbe properly mapped, and ii) a front view image is needed. Moreover, this methodis not fully applicable because of two reasons. First, its long computation timeis an important drawback. Second, the need of separately applying two differentnetworks is a relevant flaw. Each network must deal with the upper skull contourand the front view cranial features, respectively. The superimposition found bythe first network can be disrupted by that one achieved by the second network.

—On the other hand, Nickerson et al. [1991] proposed a novel methodology to findthe optimal fit between a 3D skull model and a 2D digital facial photograph. Themost important novelty of this technique was the automatic calculation of theoverlay of the skull surface mesh on the digital facial photograph. This mapping





was achieved from the matching of four landmarks previously identified both inthe face and the skull. The landmarks used in their work were: either the glabe-

lla or nasion landmarks, the two ectocanthion points, and an upper mandibulardentition point, if present, or the subnasal point. The mappings were developedfrom sets of similarity transformations and a perspective projection. The para-meters of the transformations and the projection that overlay the 3D skull onthe 2D photograph are optimized with three different methods: a heuristic, aclassic nonlinear optimization, and a binary-coded genetic algorithm, with thelatter achieving the best results.

—Ballerini et al. [2007] proposed an improvement of Nickerson et al.’s approach.The forensic experts extract different landmarks on the skull 3D model obtainedin the first stage (see Section 4.1) and on the face photograph. Then, a geneticalgorithm is used to find the optimal transformation to match them. The maindifferences between this approach and the previous one are the use of a real

coding scheme and a better design of the genetic algorithm components. Themethod for the superimposition of the 3D skull model on the 2D face photographis fully automatic.

—Ibanez et al. [2009a] extended the initial results in [Ballerini et al. 2007] to ac-complish a broader study in order to demonstrate that real-coded evolutionaryalgorithms are suitable approaches for craniofacial superimposition. In particu-lar, the authors highlight the good performance and high robustness of the stateof the art covariance matrix adaptation evolution strategy (CMA-ES) [Hansenand Ostermeier 2001]. Moreover, CMA-ES computation time is less than 15 se-conds, in the six real-world identification cases considered. It is an impressiveimprovement with respect to the manual superimposition performed by a foren-sic expert which took 24 hours. An example of the manual and computer-basedcraniofacial superimposition results is shown in Figure 7. This is a real case thathas been previously solved by the staff of the Physical Anthropology Lab at theUniversity of Granada in collaboration with the Spanish scientific police.

—Ibanez et al. [2008; 2009b] extended the approach in [Ballerini et al. 2007] consi-dering the uncertainty involved in the location of the cephalometric landmarks.In particular, authors use fuzzy logic to model the extremely difficult task of locating the landmarks [Richtsmeier et al. 1995] in a invariable place, with theaccuracy needed by craniofacial superimposition.

—Santamarıa et al. [009b] avoid the coplanarity problem that is typical in computervision by using an extended set of cephalometric landmarks including fuzzy land-marks. Those fuzzy landmarks are regions in the face image that are providedby the forensic anthropologists when it is not possible to determine an accuratelocation for the cephalometric landmarks.

4.3 Decision making

Once the skull-face overlay is achieved, the decision making stage can be tackled.The straightforward approach would involve measuring the distances between everypair of landmarks in the face and in the skull. Nevertheless, this is not advisablebecause errors are prone to be accumulated during the process of calibrating thesize of the images. Instead, studies based on proportions among landmarks are





Fig. 7. From left to right, manual and computer-aided craniofacial superimposition

preferred. Geometric figures like triangles or squares are good choices. It is alsoimportant to consider as many landmarks as possible, and different proportionsamong them [George 1993].

Although the methods described in the following are usually called in the lite-rature as computer-aided craniofacial superimposition, we prefer to refer to themas decision making methods since we think the authors fail on specifying the rightcraniofacial superimposition stage where their works are included. Indeed, the pro-posed automatic techniques are mainly focused on the decision making strategy

as they are actually decision support systems assisting the anthropologist to takethe final identification decision6. These algorithms are applied on the digitalizedimages stored on the computer, after the determination of the orientation and sizeof the skull by “routine” skull-face overlay techniques.

Tao [1986] developed the first system in which a computer was used for the de-cision making stage. That decision support system aimed to replace the previouslyused methods based on range estimation and subjective judgment. The systemprovided an identification conclusion by using distances between landmarks fromthe superimposed images. Later, Lan et al. [1988; 1990; 1993] proposed the use of 52 different superimposition identification indexes for that aim in the TLGA-213system. Those indexes were based on anthropometrical measures of Chinese adults,male and females, and were used together with proportion and distances betweensuperimposed landmarks lines to automatically compute the final identification de-

cision.Pesce Delfino et al. [1986; 1993] applied k-th-order polynomial functions and

Fourier harmonic analysis to assess the fit between the outline of the skull and

6We should remark that, although the reviewed systems are labeled as automatic in the sense thatthey are able to provide an identification decision without the intervention of the forensic expert,the supervision and final validation of the latter is always required as in any computer-aidedmedical diagnosis system [Berner 2007].





the face. Ten cases including positive and negative identifications were investi-gated. The polynomial function was used to smooth the curve representing the

investigated profile. The square root of the mean square error is taken to calculatethe distance between polynomial function curves obtained for the skull and theface profile. The Fourier analysis considered the profile as an irregular periodicfunction whose sinusoidal contributors are found. Low-order harmonics (the firstthree or four) represented the basic profile shape and the high order harmonicscorresponded to details. The sum of the amplitude differences of the sinusoidalcontributors between profiles of the skull and the face represented the second inde-pendent parameter for numerical comparison. A Janus procedure (so called by theauthors because of the double-headed Latin god Janus, the bi-front) was used toevaluate the symmetry differences between the two profiles. This procedure takesinto account the relationship between the total arc and the chord length and thearea they delimit in the two faced profiles. All these parameters are calculated by

a computer software package called Shape Analytic Morphometry. However, thismethod would be only applicable when lateral or oblique photographs are availa-ble. Meanwhile, their contribution requires manual repositioning of the skull forthe correct superimposition.

Bajnoczky and Kiralyfalvi [1995] used the difference between the coordinate va-lues of the pair of anatomical and/or anthropometrical points in both skull and facefor judging the match between the skull and facial image obtained by the super-imposition technique. Eight to twelve pairs of points were recorded and expressedas pixel units. Then, the final matrix, containing coordinates of measured pointsand calculated values, was established by computer-aided processing. Lacking theappropriate information, their model assumed that all data in that matrix wereindependent and followed a normal distribution with the same variance. A part of that variance was σ2, which was the square of the measurement error and was itself assumed to be the same with all the data. The model of the authors was based onassumptions that take the form:

The components of the error term are independent and distributed according to N (0, 2 σ2). (1)

The authors used a presupposed value of σ as part of the model assumptions.Under the assumption that the null hypothesis (Equation 1) is valid, it was statis-tically tested using two values for σ. Authors claimed that when a given case isevaluated it is crucial to know what value can be considered as measurement error.One skull and two photographs were used to test the method. Both frontal andlateral face photographs are considered. They noted that their method is suitablefor filtering out false positive identifications. Although the results obtained fromthis method are objective and easily interpreted for lay people, the anatomical andanthropometrical consistency between the skull and the face should be assessed byforensic examiners who are well versed in the anatomy of the skull and face. Theauthors conclude that their method should be used only in combination with classicvideo superimposition and could be regarded as an independent check.

In Yoshino et al.’s skull identification system [Yoshino et al. 1997] the distancebetween the landmarks and the thickness of the soft tissue of the anthropometricalpoints are semi-automatically measured on the monitor for the assessment of the





anatomical consistency between the digitalized skull and face. The consistency isbased on 13 criteria they previously defined using 52 skulls [Yoshino et al. 1995].

The software includes polynomial functions and Fourier harmonic analysis for eva-luating the match of the outline such as the forehead and mandibular line in bothdigitized images. To extract the outline, gradient and threshold operations are used.Five case studies are carried out. However, they noted that these analysis could notalways be applied because of the difficulties in extracting the facial contour fromsmall and poor facial photographs offered from the victim’s family.

The skull-face overlay in Ricci et al. [2006] was guided by different crosses thatwere manually marked by the human operator in both the face and the skull ra-diograph photographs. Once that stage was tackled, the algorithm calculated thedistance of each cross moved and the respective mean in pixels. The algorithmconsidered a 7-pixel distance a negligible move. The mean value of the total dis-tance in crosses moved represented the index of similarity between the given face

and skull: the smaller the index value, the greater the similarity. The algorithmsuggested an identification decision based on that index of similarity. The authorsclaim 100% of correct identification over 196 cross-comparisons and report that theminimal number of needed landmarks is 4.

5. RELATED WORKS

Nearly all the methods described in the previous section use anthropological land-marks to compute and/or to assess the fit between the skull and the face, but wefound only one paper that addresses their automatic extraction in the skull [Parzia-nello et al. 1996]. The authors proposed a method based on simple image proces-sing algorithms for the detection of craniometric points in video-based skull images.Their method works on 2D digitalized images of undamaged skulls and assumes they

are in frontal view. The authors claimed that their method produces good resultsand it is potentially useful for the craniofacial superimposition process. However,we did not find any paper describing a system using the automatically extractedlandmarks.

The literature on facial feature extraction is huge. Douglas [2004] reviews a num-ber of image processing algorithms for the automatic extraction of landmarks inphotograms and cephalograms. Interesting algorithms combining artificial intelli-gence techniques have been successfully developed and applied. However they areout of the scope of this survey as well, as they are related to studies on craniofacialsurgery or on face recognition and not to forensic skull identification.

Readers interested in 3D cranial landmark categorization can refer to [Brownet al. 2004]. The accurate placement of anatomical features for craniofacial super-

imposition is a real need nowadays [Stephan 2009c]. However, to the best of ourknowledge, no method for their automatic localization is found in the literature.The area of 3D face processing [Zhen and Huang 2004; Zhao and Chellapa 2005]

could also seem to have some relation with craniofacial superimposition, specificallythe face modeling topic. 3D face processing methods deal with the very complextask of properly turning a 3D object (the subject face) into a 2D image. Obtaininga skull 3D model is feasible –as well as very useful to improve the identificationprocess– due to the availability of the physical object in the forensic anthropology





lab (see Section 4.1). Nevertheless, the existing powerful methods in 3D face mode-ling (such as [Shan et al. 2001]) are not applied since craniofacial superimposition

deals with the identification of deceased people. Hence, it is usually difficult for theanthropologist to get significant data in real conditions to apply the latter techni-ques. The availability of photographs and videos of the sample of candidates is low.This is one of the reasons why the currently established fundamentals of the foren-sic technique are based either on a 2D skull photo-2D face photo or on a 3D skullmodel-2D face photo comparison. The use of 3D facial models is not consideredby forensic anthropologists nowadays although it could become a challenge in thefuture when the massive use of video and imaging devices the world is experiencingwill solve the problem of the lack of subject data.

There are some other approaches apart from craniofacial superimposition puttinginto correspondence 3D face models and 2D face photographs. On the one hand,there is the well known area of face recognition [Zhao et al. 2003] which includes

several multimodal 3D-2D approaches [Bowyer et al. 2006]. One of the most repre-sentative works in this subarea is that by Blanz and Vetter [2003]. On the otherhand, we can find techniques tackling 3D face model-2D face photograph super-imposition for personal identification [Goos et al. 2006; De Angelis et al. 2009].The opposite problem, the registration of 2D face photos to a 3D facial model, isconsidered in [Clarkson et al. 2001]. Of course, all the latter approaches are out of the scope of this survey as they deal with completely different problems and data(3D face models instead of 3D skull models).

Besides, recent literature has considerably developed the potential of a somehowrelated task to craniofacial superimposition, craniofacial reconstruction, preferablyreferred to as facial approximation [Vanezis et al. 2000; Claes et al. 2006; Wilkinson2009b]. The 3D facial image may be reconstructed by either building muscle andsoft tissue using clay, or by means of computer graphics. Data concerning the

reliability of these methods for forensic anthropology and the lack of relationshipbetween facial approximation and resemblance rating have been reported [Stephanand Arthur 2006]. An interesting review of current systems for computer-aidedforensic facial reconstruction can be found in [Wilkinson 2005].

6. DISCUSSION AND RECOMMENDATIONS FOR FUTURE RESEARCH

In this final section, we summarize the results by reporting a number of solvedand unsolved problems in craniofacial superimposition as well as possibilities forforensic applications by identifying some trends in the field. We also provide a listof recommendations for future research. Finally we talk about the need of publicforensic data and we present the web site we have created for this purpose.

6.1 Solved and unsolved problems

To date, no fully automatic method is used in practical applications despite thehigh number of cases examined [Ubelaker 2000] and the large amount of time theforensic experts need to spend in performing such examination.

All the researchers agree that a key problem of craniofacial superimposition isthe size and orientation of the skull to correctly match that one of the head in thephoto. In fact, it would be very difficult to obtain the technical specifications usedto take the antemortem photograph.





The need of a sophisticated procedure or an expensive hardware configuration toimplement a digitalized 3D cranial image reconstruction has been stated to be the

reason why computer-aided automatic craniofacial superimposition methods didnot gain much popularity [Ghosh and Sinha 2001]. Nevertheless, the acquisitionof a 3D model of the skull should not be a hinder nowadays. Indeed, such modelcould be reconstructed either by the scanner’s software, when the rotary device isavailable, or by range image registration algorithms (Santamarıa et al.[007a; 009a]).

Regardless the automatic or non automatic nature of the approach to tackle thecraniofacial superimposition, some authors [Cattaneo 2007; Yoshino et al. 1997;Jayaprakash et al. 2001; Shahrom et al. 1996] agree that this technique shouldbe used only for excluding identity, rather than for positive identification. [Setaand Yoshino 1993] state the general rule that superimposition is of greater value inruling out a match, because it can be definitely stated that the skull and photographare not those of the same person. However, if they do align, it can only be stated

that the skull might possibly be that of the person in the photograph. On theother hand, a research carried out on very large number of comparisons indicatesthat there is a 9% chance of misidentification if just one photograph is used for thecomparison, and this probability of false identification diminishes to less than 1% if multiple photographs from widely different angles to the camera are used [Austin-Smith and Maples 1994].

Skull identification by craniofacial superimposition peaked in the 1990-1994period and subsequently declined, with last use of the technique occurring in1996 [Ubelaker 2000]. According to Ubelaker, these frequencies appear to reflectthe availability of the necessary equipment and expertise in 1990, coupled withawareness of the value of this approach in the forensic science and law enforcementcommunities. The decline in use likely reflects both the increased awareness of the

limitations of this technique and the greater availability of more precise methodsof identification, especially the molecular approaches [Ubelaker 2000].

We have to stress that these statements were valid when the equipments wereeither very expensive or not very accurate. Nowadays, the limitations pointed outby some researchers, like the poor quality of the antemortem photographs [Ubelakeret al. 1992; Nickerson et al. 1991] or the curved surface of the monitor [Shahromet al. 1996], which were claimed to influence a correct superimposition, should notbe a problem.

We believe that most of the claimed difficulties in finding an accurate magni-fication and orientation of skull can be currently overcome by a computer-aidedautomatic craniofacial superimposition method. Meanwhile, other reasons addu-ced for a scarce reliability in such methods are definitively overcome nowadays. Inparticular, the high computation time Nickerson et al. [1991] required several daysto achieve an automatic craniofacial superimposition).

Plenty of research has also been focused on supporting the anthropologist inthe decision making stage [Pesce Delfino et al. 1986; Yoshino et al. 1995; Ricciet al. 2006] and/or the validity of “routine” superimposition methods [Bajnoczkyand Kiralyfalvi 1995; Scully and Nambiar 2002]. At the same time, Jayaprakashet al. [2001] affirmed that visual assessment is more effective than metrical studies.Indeed, the method they propose, called “craniofacial morphoanalysis” is based on





the visual evaluation of a number of attributes. This is the reason why this methodis not included in the computer-aided approaches described in this survey.

6.2 Trends

Video superimposition has been preferred to photographic superimposition sincethe former is simpler and quicker [Jayaprakash et al. 2001]. Video superimpositionovercomes the protracted time involved with photographic superimposition, wheremany photographs of the skull had to be taken in varying orientations [Nickersonet al. 1991]. However, it has been indicated that craniofacial superimposition basedon the use of photographs is better than using video in terms of resolutions of details [Yoshino et al. 1995]. On the other hand, the fade and wipe facility in videosuperimposition allows the expert to analyze the congruence in every sector of thesuperimposed images, thereby rendering this method more popular [Jayaprakashet al. 2001]. Nevertheless, this process is still quite subjective, relying on the skill

and dexterity of the operator [Nickerson et al. 1991].Recent papers confirm that some authors think the most advanced method is ba-sed on computer-aided craniofacial superimposition through the use of the imagingtools provided by Adobe PhotoshopTMand Corel DrawTMsoftware packages [Al-Amad et al. 2006; Bilge et al. 2003; Ross 2004]. We agree with these authors thatworking with digital images is definitively simpler and cheaper than with photo-graphic or video superimposition equipments. However, we should note that themethods they use are not automatic as they manually resize, shift and rotate theimages by trial and error. Thus, they deal with a very time consuming and erroraffected process7.

There seems to be an increasing interest in facial reconstruction or approximation.Besides the advantages and disadvantages described in Wilkinson’s review [Wilkin-son 2005], and the undoubted attractiveness of these techniques, we believe that

they still need extensive research before being fully accepted in forensic investiga-tions. Indeed, researchers involved in facial approximation think that craniofacialsuperimposition may be preferred to reconstruction in cases where some clues canlimit the identity to a few candidates [Turner et al. 2005].

6.3 Recommendations

It would be worthwhile to investigate how all the manual steps of the routinemethods described in Section 4, from the skull modeling to the decision making,can be automated.

Automatic localization of anthropological landmarks on 3D skull model and 2Dface images are few examples of useful potential applications of image processingtechniques to forensic sciences. Computer graphics techniques can provide accurate

and automatic registration methods for 3D model building and for superimpositionof 3D models on 2D images, which are a real need.

The use of 3D models of skulls should be preferred to their 2D representation(like photographs [Ghosh and Sinha 2001] or radiographs [Ricci et al. 2006]) due to

7It is worth ro remind that the forensic expert employed approximately 24 hours to manuallysuperimpose the skull and the face shown in Figure 7 (left) following a similar computer-aidedprocedure.





the inherent problems of representing a 3D object with a 2D image.Besides, we can draw the underlying uncertainty involved in the craniofacial

superimposition process. The correspondence between facial and cranial anthropo-metric landmarks is not always symmetrical and perpendicular: some landmarksare located in a higher position in the alive person face than in the skull, and someothers have not got a directly related landmark in the other set. The identificationdecision is to be expressed according several confidence levels (“absolute matching”,“absolute mismatching”, “relative matching”, “relative mismatching”, and “lack of information”). Hence, we again find the uncertainty and partial truth involved inthe identification process. In conclusion, fuzzy logic could be an interesting tool tobe applied. However, we have only found few proposals considering this artificialintelligence tool [Ghosh and Sinha 2001], Ibanez et al. [2008; 2009b].

It is recommended to use recent photographs or not to consider age-related fea-tures, otherwise algorithms for predicting what an adult head and face at one point

in time might look like several years later will be necessary [Albert et al. 2007].The distortions that may arise during the craniofacial superimposition processcould influence the reliability of the identification. It is advised to use centralprojections or to apply a mathematical model to eliminate the distortions [Eliasovaand Krsek 2007].

In computer-aided diagnosis, the general agreement is that the focus should beon making useful computer-generated information available to physicians for deci-sion support rather than trying to make a computer act like a diagnostician [Berner2007]. Following the same track, the final goal of computer-aided automatic cranio-facial identification systems should be to provide the forensic anthropologists withidentification decisions they only have to supervise and validate.

6.4 The craniofacial superimposition challenge

Science evolves thanks to the knowledge exchange and the chance to either improveexisting approaches or propose new methods for the problems that are tackled. The-refore, it is essential to guarantee objective procedures to evaluate the performanceof those proposals.

Unlike other related research fields like face recognition or machine learning, it isnot possible to compare the performance of the developed craniofacial superimpo-sition methods since there is not a common forensic dataset available comprised by3D partial views of skulls, the corresponding reconstructed 3D model, photographsof the person the skull belongs to, landmarks, superimposition results detailing theused techniques together with the identification decision and, so on. This fact hasalready been mentioned by some experts on the area, such as Carl N. Stephan (see[2009a] and his sentence quoted in the Introduction section).

We think this is the main reason for finding few practical applications of computer-aided automatic methods. In our opinion, having forensic material avai-lable is a keystone for the advance of the craniofacial superimposition research field.

Large, publicly available databases of known case studies should be collected.Those databases will encourage development and testing of new methods. Theywill also allow the validation of the methods by applying them to solved cases andby comparing the results with the identification previously determined by forensicanthropologists.





Assuming this challenge, we have created a web site8 with the aim to provideforensic data to the research community and to join forces by the collaboration of

other forensic labs. Of course, legal aspects have been taken into account.As said, there are different issues in craniofacial superimposition that can be

tackled by means of advanced artificial intelligence approaches. Evolutionary al-gorithms, fuzzy logic, and neural networks have demonstrated their suitability fortackling different craniofacial superimposition tasks. Moreover, the application of these techniques to the craniofacial superimposition problem has been presented inthis survey as an emerging trend. Thus, public craniofacial superimposition data-sets will be specially interesting for the artificial intelligence research community.Indeed, different authors have recently claimed that a multidisciplinary researchteam is a real need in forensic identification by craniofacial superimposition nowa-days [Ricci et al. 2006; Benazzi et al. 2009].

REFERENCES

Al-Amad, S., McCullough, M., Graham, J., Clement, J., and Hill, A. 2006. Craniofacialidentification by computer-mediated superimposition. Journal of Forensic Odontostomato-logy 24, 47–52.

Albert, A. M., Ricanek Jr., K., and Patterson, E. 2007. A review of the literature on theaging adult skull and face: Implications for forensic science research and applications. Forensic Science International 172 , 1–9.

Aleman, I., Botella, M. C., and Ruız, L. 1997. Determinacion del sexo en el esqueleto post-craneal. Estudio de una poblacion mediterranea actual (in Spanish). Archivo Espa˜ nol de Mor- fologıa 2 , 69–79.

Aulsebrook, W. A., I.-M. Y., Slabbert, J. H., and Becker, P. 1995. Superimposition andreconstruction in forensic facial identification: a survey. Forensic Science International 75, 2-3(October), 101–120.

Austin-Smith, D. and Maples, W. R. 1994. The reliability of skull/photograph superimpositionin individual identification. Journal of Forensic Sciences 39, 2 (March), 446–455.

Back, T., Fogel, D. B., and Michalewicz, Z., Eds. 1997. Handbook of Evolutionary Compu-tation . IOP Publishing Ltd and Oxford University Press, Bristol, UK.

Bajnoczky, I. and Kiralyfalvi, L. 1995. A new approach to computer-aided comparison of skull and photograph. International Journal of Legal Medicine 108 , 157–161.

Ballerini, L., Cordon, O., Damas, S., and Santamarıa, J. 2009. Automatic 3D modeling of skulls by scatter search and heuristic features. In Applications of Soft Computing. Updating the State of the Art , E. Avineri, M. Koepen, K. Dahal, Y. Sunitiyoso, and R. Roy, Eds. Springer,149–158.

Ballerini, L., Cordon, O., Damas, S., Santamarıa, J., Aleman, I., and Botella, M. 2007.Craniofacial superimposition in forensic identification using genetic algorithms. In Proceedingsof the IEEE International Workshop on Computational Forensics. Springer, Manchester, UK,429–434.

Benazzi, S., Fantini, M., De Crescenzio F.and Mallegni, G., Mallegni, F., Persiani, F.,and Gruppioni, G. 2009. The face of the poet dante alighieri reconstructed by virtual modelling

and forensic anthropology techniques. Journal of Archaeological Science 36, 2, 278–283.Bernardini, F. and Rushmeier, H. 2002. The 3D model acquisition pipeline. Computer Graphics

Forum 21, 2, 149–172.

Berner, E. S., Ed. 2007. Clinical Decision Support Systems: Theory and Practice . Springer,New York, USA.

Bertillon, A. 1896. The Bertillon system of identification. Nature 54, 1407 (October), 569–570.

8http://www.softcomputing.es/socovifi/en/home.php





Besl, P. and McKay, N. 1992. A method for registration of 3-D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence 14, 2 (February), 239–256.

Bilge, Y., Kedici, P., Alakoc, Y., K., U., and Ilkyaz, Y. 2003. The identification of a dis-membered human body: a multidisciplinary approach. Forensic Science International 137, 2–3,141–146.

Biwasaka, H., Saigusa, K., and Aoki, Y. 2005. The applicability of holography in forensicidentification: a fusion of the traditional optical technique and digital technique. Journal of Forensic Sciences 50, 2, 393–399.

Blanz, V. and Vetter, T. 2003. Face recognition based on fitting a 3D morphable model. IEEE Transactions on Pattern Analysis and Machine Intelligence 25, 9, 1–12.

Bowyer, K. W., Chang, K., and Flynn, P. 2006. A survey of approaches and challenges in 3Dand multi-modal 3D + 2D face recognition. Computer Vision and Image Understanding 101, 1(January).

Broca, P. 1875. Instructions craniologiques et craniometriques (in french). Memoires de la Societe d’Anthropologie de Paris, 63–96.

Brocklebank, L. M. and Holmgren, C. J. 1989. Development of equipment for the standardisa-tion of skull photographs in personal identifications by photographic superimposition. Journal of Forensic Sciences 34, 5 (September), 1214–1221.

Bronkhorst, D. 2006. Truth and justice: A Guide to Truth Commissions and TransitionalJustice. Amnesty International Dutch Section. 2nd edition.

Brown, L. 1992. A survey of image registration techniques. ACM Computing Surveys 24, 4(December), 325–376.

Brown, R. E., Kelliher, T. P., Tu, P. H., Turner, W. D., Taister, M. A., and Miller,

K. W. P. 2004. A survey of tissue-depth landmarks for facial approximation. Forensic Science Communications 6, 1, [online].

Burns, K. 2007. Forensic Anthropology Training Manual . Prentice-Hall.

Cattaneo, C. 2007. Forensic anthropology: development of a classical discipline in the newmillennium. Forensic Science International 165, 2-3 (January), 185–193.

Chandra Sekharan, P. 1993. Positioning the skull for superimposition. In Forensic Analysis of the Skull , M. Y. Iscan and R. Helmer, Eds. Wiley, New York, USA, 105–118.

Claes, P., Vandermeulen, D., De Greef, S., Willems, G., and Suetens, P. 2006. Cranio-

facial reconstruction using a combined statistical model of face shape and soft tissue depths:methodology and validation. Forensic Science International 159S , S147–S158.

Clarkson, M. J., Rueckert, D., Hill, D. L. G., and Hawkes, D. J. 2001. Using photo-consistency to register 2D optical images of the human face to a 3D surface model. IEEE Transactions on Pattern Analysis and Machine Intelligence 23, 11 (November), 1266–1280.

Clement, J. G. and Ranson, D. L. 1998. Craniofacial Identification in Forensic Medicine .Oxford University Press, New York, USA.

De Angelis, D., Sala, R., Cantatore, A., Grandi, M., and Cattaneo, C. 2009. A newcomputer-assisted technique to aid personal identification. International Journal of Legal Me-dicine 123, 4 (July), 351–356.

Dorion, R. B. 1983. Photographic superimposition. Journal of Forensic Sciences 28, 3 (July),724–734.

Douglas, T. S. 2004. Image processing for craniofacial landmark identification and measure-ment: a review of photogrammetry and cephalometry. Computerized Medical Imaging and

Graphics 28, 7 (October).Eliasova, H. and Krsek, P. 2007. Superimposition and projective transformation of 3D object.

Forensic Science International 167, 2–3 (April), 146–153.

Enciso, R., Memon, A., and Mah, J. 2003. Three-dimensional visualization of the craniofacialpatient: volume segmentation, data integration and animation. Orthodontics & Craniofacial Research 6, s1, 66–71.

Fantini, M., De Crescenzio, F., Persiani, F., and Benazzi, S. 2008. 3D restitution, restorationand prototyping of a medieval damaged skull. Rapid Prototyping Journal 14, 5 (September),1–1.





Fenton, T. W., Heard, A. N., and Sauer, N. J. 2008. Skull-photo superimposition and bor-der deaths: identification through exclusion and the failure to exclude. Journal of Forensic

Sciences 53, 1 (January), 34–40.Galantucci, L., Percoco, G., Angelelli, G.and Lopez, C., Introna, F., Liuzzi, C., and

De Donno, A. 2006. Reverse engineering techniques applied to a human skull, for cad 3Dreconstruction and physical replication by rapid prototyping. Journal of Medical Engineering & Technology 30, 2 (April), 102–111.

George, R. M. 1993. Anatomical and artistic guidelines for forensic facial reconstruction. InForensic Analysis of the Skull , M. Y. Iscan and R. Helmer, Eds. Wiley, New York, USA, 215–227.

Ghosh, A. and Sinha, P. 2001. An economised craniofacial identification system. Forensic Science International 117, 1-2 (March), 109–119.

Ghosh, A. and Sinha, P. 2005. An unusual case of cranial image recognition. Forensic Science International 148, 2-3 (March), 93–100.

Glaister, J. and Brash, J. 1937. Medico-Legal Aspects of the Ruxton Case . E. & S. Livingstone,Edinburgh, U.K.

Gonzalez, R. C. and Woods, R. E. 2008. Digital Image Processing . Addison-Wesley. 3rdedition.

Gonzalez-Colmenares, G., B.-L. M. M.-R. G. F.-C. J. 2007. Age estimation by a dentalmetohd: A comparison of lamedin’s and prince & ubelaker’s technique. Journal of Forensic Sciences 52, 5, 1156–1160.

Goos, M. I., Alberink, I. B., and Ruifrok, A. C. 2006. 2D/3D image (facial) comparison usingcamera matching. Forensic Science International 163, 1–2 (December), 10–17.

Hansen, N. and Ostermeier, A. 2001. Completely derandomized self-adaptation in evolutionstrategies. Evolutionary Computation 9, 2, 159–195.

Helmer, R. 1988. Identifizierung der leichenuberreste das josef mengele (in German). In Thesiscollection of international symposium on skull-image superimposition , R. Helmet, Ed. Kiel,Germany, 3.

Ibanez, O., Cordon, O., Damas, S., and Santamarıa, J. 2008. Craniofacial superimpositionbased on genetic algorithms and fuzzy location of cephalometric landmarks. In Hybrid Artificial

Intelligence Systems. Number 5271 in Lecture Notes in Artificial Intelligence. Springer, Berlin,Germany, 599–608.

Ibanez, O., Cordon, O., Damas, S., and Santamarıa, J. 2009a. An experimental study on theapplicability of evolutionary algorithms to craniofacial superimposition in forensic identification.Information Sciences. In press.

Ibanez, O., Cordon, O., Damas, S., and Santamarıa, J. 2009b. A new approach to fuzzylocation of cephalometric landmarks in craniofacial superimposition. In International Fuzzy Systems Association - European Society for Fuzzy Logic and technologies (IFSA-EUSFLAT)World Congress. Lisbon, Portugal, 195–200.

Ikeuchi, K. and Sato, Y. 2001. Modeling from Reality . Kluwer Academic Publishers, Norwell,USA.

Indriati, E. 2009. Historical perspectives on forensic anthropology in Indonesia. In Handbook of Forensic Anthropology and Archaeology , S. Blau and D. H. Ubelaker, Eds. Left Coast Press,California, USA, 115–125.

Iscan, M. Y. 1993. Introduction to techniques for photographic comparison. In Forensic Analysisof the Skull , M. Y. Iscan and R. Helmer, Eds. Wiley, New York, USA, 57–90.

Iscan, M. Y. 2005. Forensic anthropology of sex and body size. Forensic Science Internatio-nal 147 , 107–112.

Jayaprakash, P. T., Srinivasan, G. J., and Amravaneswaran, M. G. 2001. Craniofacial morp-hoanalysis: a new method for enhancing reliability while identifying skulls by photosuperimpo-sition. Forensic Science International 117 , 121–143.

Keen, P. G. W., Ed. 1978. Decision Support Systems: an Organizational Perspective . Addison-Wesley Pub. Co., Reading, Mass.





Krogman, W. M. and Iscan, M. Y. 1986. The Human Skeleton in Forensic Medicine . CharlesC. Thomas, Springfield, USA. 2nd edition.

Kumari, T. R. and Chandra Sekharan, P. 1992. Remote control skull positioning device forsuperimposition studies. Forensic Science International 54, 127–133.

Laguna, M. and Martı, R. 2003. Scatter Search: Methodology and Implementations in C . KluwerAcademic Publishers, Boston, , USA.

Lan, Y. 1990. Research report on model tlga-213 image superimposition identification system.In Spec Issue Criminal Technol Supplement . The Fifth Bureau of the National Public SecurityDepartment. Beijing, China, 13.

Lan, Y. 1992. Development and current status of skull image superimposition methodology andinstrumentation. Forensic Science Review 4, 2, 126–136.

Lan, Y. and Cai, D. 1985. Study on model tlga-1 skull identification apparatus. In Spec Issue Criminal Technol Supplement . The Fifth Bureau of the National Public Security Department.Beijing, China, Germany, 23.

Lan, Y. and Cai, D. 1988. A new technology in skull identification. In Advances in Skull Identification Via Video Superimposition , R. Helmet, Ed. Kiel, Germany, 3.

Lan, Y. and Cai, D. 1993. Technical advances in skull-to-photo superimposition. In Forensic Analysis of the Skull , M. Y. Iscan and R. Helmer, Eds. Wiley, New York, USA, 119–129.

Maat, G. J. R. 1989. The positioning and magnification of faces and skulls for photographicsuperimposition. Forensic Science International 41, 3, 225–235.

Martin, R. and Saller, K. 1966. Lehrbuch der Anthropologie in Systematischer Darstellung (in German). Gustav Fischer Verlag, Stuttgart.

Mitchell, T. 1997. Machine Learning . McGraw Hill, New York, USA.

Nakasima, A., T. M.-M. N. H.-Y. T. K.-A. Y. H. S. 2005. Three-dimensional computer-generated head model reconstructed from cephalograms, facial photographs, and dental castmodels. American Journal of Orthodontics and Dentofacial Orthopedics 127, 3 (March), 282–292.

Nickerson, B. A., Fitzhorn, P. A., Koch, S. K., and Charney, M. 1991. A methodology fornear-optimal computational superimposition of two-dimensional digital facial photographs andthree-dimensional cranial surface meshes. Journal of Forensic Sciences 36, 2 (March), 480–500.

Parzianello, L. C., Da Silveira, M. A. M., Furuie, S. S., and Palhares, F. A. B. 1996.

Automatic detection of the craniometric p oints for craniofacial identification. In Anais do IX SIBGRAPI’96 . Sociedade Brasileira de Computacao, Caxambu, Brazil, 189–196.

Pesce Delfino, V., Colonna, M., Vacca, E., Potente, F., and Introna Jr., F. 1986.Computer-aided skull/face superimposition. American Journal of Forensic Medicine and Pat-hology 7, 3, 201–212.

Pesce Delfino, V., Vacca, E., Potente, F., Lettini, T., and Colonna, M. 1993. Shapeanalytical morphometry in computer-aided skull identification via video superimposition. InForensic Analysis of the Skull , M. Y. Iscan and R. Helmer, Eds. Wiley, New York, USA, 131–159.

Pickering, R. and Bachman, D. 2009. The Use of Forensic Anthropology . CRC Press, NewYork, USA. 2nd edition.

Ranson, D. 2009. Legal aspects of identification. In Handbook of Forensic Anthropology and Archaeology , S. Blau and D. H. Ubelaker, Eds. Left Coast Press, California, USA.

Ricci, A., Marella, G. L., and Apostol, M. A. 2006. A new experimental approach to

computer-aided face/skull identification in forensic anthropology. American Journal of Fo-rensic Medicine and Pathology 27, 1 (March), 46–49.

Richtsmeier, J., Paik, P., Elfert, P., Cole, T., and Dahlman, F. 1995. Precision, repeatabilityand validation of the localization of cranial landmarks using computed tomography scans. The Cleft Palate-Craniofacial Journal 32, 3, 217–227.

Ross, A. H. 2004. Use of digital imaging in the identification of fragmentary human skeletalremains: A case from the Republic of Panama. Forensic Science Communications 6, 4, [online].

Rumelhart, D. E. and McClelland, D. 1986. Parallel Distributed Processing: Explorations in the Microstructure of Cognition . MIT Press, Massachusetts, USA.





Santamarıa, J., Cordon, O., and Damas, S. 2007a. A scatter search-based technique for pair-wise 3D range image registration in forensic anthropology. Soft Computing 11, 9, 819–828.

Santamarıa, J., Cordon, O., Damas, S., Aleman, I., and Botella, M. 2007b. Evolutionaryapproaches for automatic 3D modeling of skulls in forensic identification. In Applications of evolutionary computing . Number 4448 in Lecture Notes in Computer Science. Springer, Berlin,Germany, 415–422.

Santamarıa, J., Cordon, O., Damas, S., Garcıa-Torres, J. M., and Quirin, A. 2009a. Per-formance evaluation of memetic approaches in 3D reconstruction of forensic objects. Soft Computing 13, 8-9, 883–904.

Santamarıa, J., Cordon, O., Damas, S., and Ibanez, O. 2009b. Tackling the coplanarityproblem in 3D camera calibration by means of fuzzy landmarks: a performance study in forensiccraniofacial superimposition. In IEEE International Conference on Computer Vision . Kyoto,Japan. In press.

Scully, B. and Nambiar, P. 2002. Determining the validity of Furue’s method of craniofacialsuperimposition for identification. Malaysian Journal of Computer Science 9, 1, 17–22.

Sen, N. K. 1962. Identification by superimposed photographs. International Criminal Police Review 162 , 284–286.

Seta, S. and Yoshino, M. 1993. A combined apparatus for photographic and video superimpo-sition. In Forensic Analysis of the Skull , M. Y. Iscan and R. Helmer, Eds. Wiley, New York,USA, 161–169.

Shahrom, A. W., Vanezis, P., Chapman, R. C., Gonzales, A., Blenkinsop, C., and Rossi,

M. L. 1996. Techniques in facial identification: computer-aided facial reconstruction usinga laser scanner and video superimposition. International Journal of Legal Medicine 108, 4,194–200.

Shan, Y., Liu, Z., and Z., Z. 2001. Model-based bundle adjustment with application to facemodeling. In IEEE International Conference on Computer Vision . Vol. 2. Vancouver, Canada,644–651.

Singare, S., Lian, Q., Wang, W. P., Wang, J., Liu, Y., Li, D., and Lu, B. 2009. Rapid prototy-ping assisted surgery planning and custom implant design. Rapid Prototyping Journal 15, 1,19–23.

Sinha, P. 1998. A symmetry perceiving adaptive neural network and facial image recognition.

Forensic Science International 98, 1-2 (November), 67–89.Stephan, C. N. 2009a. Craniofacial identification: method background and overview.

http://www.craniofacial.com.

Stephan, C. N. 2009b. Craniofacial identification: techniques of facial approximation and cra-niofacial superimposition. In Handbook of Forensic Anthropology and Archaeology , S. Blau andD. H. Ubelaker, Eds. Left Coast Press, California, USA, 304–321.

Stephan, C. N. and Arthur, R. S. 2006. Assessing facial approximation accuracy: How doresemblance ratings of disparate faces compare to recognition tests? Forensic Science Interna-tional 159S , S159–S163.

Stephan, C. N., Taylor, R. G., and Taylor, J. A. 2008. Methods of facial approximation andskull-face superimposition, with special consideration of method development in Australia. InForensic Approaches to Death, Disaster and Abuse , M. Oxenham, Ed. Australian AcademicPress.

Stephan, C. N., H.-A. D. P. 2009c. Further evidence on the anatomical placement of the

human eyeball for facial approximation and craniofacial superimposition. Journal of Forensic Sciences 54, 2 (March), 267–269.

Stratmann, H. 1998. Excuses for the truth. http://home.wxs.nl/˜loz/maneng.htm.

Tao, C. 1986. Report on computer programming for model tlga-1 skull identification. In Spec Issue Criminal Technol Supplement . The Fifth Bureau of the National Public Security Department.Beijing, China, 41.

Turner, W. D., Brown, R. E. B., Kelliher, T. P., Tu, P. H., Taister, M. A., and Miller,

K. W. P. 2005. A novel method of automated skull registration for forensic facial approximation.Forensic Science International 154, 149–158.





Ubelaker, D. H. 2000. A history of Smithsonian-FBI collaboration in forensic anthropology,especially in regard to facial imagery. Forensic Science Communications 2, 4, [online].

Ubelaker, D. H., Bubniak, E., and O’Donnel, G. 1992. Computer-assisted photographic su-perimposition. Journal of Forensic Sciences 37, 3 (May), 750–762.

Vanezis, P., Vanezis, M., McCombe, G., and Niblett, T. 2000. Facial reconstruction using3-D computer graphics. Forensic Science International 108 , 81–95.

Webster, G. 1955. Photography as an aid in identification: the Plumbago Pit case. Police Journal 28 , 185–191.

Welcker, H. 1867. Der schadel Dantes (in German). In Jahrbuch der deutschen Dantegesells-chaft , K. Witte and G. Boehmer, Eds. Vol. 1. Brockhaus, Liepzig, 35–56.

Wilkinson, C. 2005. Computerized forensic facial reconstruction: A review of current systems.Forensic Science, Medicine, and Pathology 1, 3, 173–177.

Wilkinson, C. 2009a. 13th meeting of International Association of Craniofacial Identification(IACI). Forensic Science, Medicine, and Pathology 5, 1 (January), 1.

Wilkinson, C. 2009b. Facial identification of the dead. Journal of Anatomy . In press.

Yoshino, M., Imaizumi, K., Miyasaka, S., and Seta, S. 1995. Evaluation of anatomical consis-

tency in craniofacial superimposition images. Forensic Science International 74, 125–134.Yoshino, M., Matsuda, H., Kubota, S., Imaizumi, K., Miyasaka, S., and Seta, S. 1997.Computer-assisted skull identification system using video superimposition. Forensic Science International 90 , 231–244.

Yoshino, M. and Seta, S. 2000. Skull-photo superimposition. In Encyclopedia of Forensic Sciences, J. A. Siegel, G. C. Knupfer, and P. J. Saukko, Eds. Vol. 2. Elsevier Science andTechnology, London, UK, 807–815.

Zadeh, L. A. 1965. Fuzzy sets. Information and Control 8 , 338–353.

Zhao, W. and Chellapa, R., Eds. 2005. Face Processing: Advanced Modeling and Methods.Elsevier.

Zhao, W., Chellappa, R., Phillips, P., and Rosenfeld, A. 2003. Face recognition: A literaturesurvey. ACM Computing Surveys 35, 4, 399–458.

Zhen, W. and Huang, T. 2004. 3D Face Processing: Modeling, Analysis and Synthesis. Springer.

Zitova, B. and Flusser, J. 2003. Image registration methods: a survey. Image and Vision Computing 21, 977–1000.

...


forensic identification computed

Documents